kern/163064: [fix][sym]driver tries to nego synch trans with
everyone
Peter
pmc at citylink.dinoex.sub.org
Sun Dec 4 18:20:09 UTC 2011
>Number: 163064
>Category: kern
>Synopsis: [fix][sym]driver tries to nego synch trans with everyone
>Confidential: no
>Severity: non-critical
>Priority: low
>Responsible: freebsd-bugs
>State: open
>Quarter:
>Keywords:
>Date-Required:
>Class: sw-bug
>Submitter-Id: current-users
>Arrival-Date: Sun Dec 04 18:20:08 UTC 2011
>Closed-Date:
>Last-Modified:
>Originator: Peter Much
>Release: FreeBSD 7.4-STABLE i386
>Organization:
n/a
>Environment:
System: FreeBSD disp.oper.dinoex.org 7.4-STABLE FreeBSD 7.4-STABLE #0: Sun Dec 4 07:09:44 CET 2011 root at disp.oper.dinoex.org:/usr/src/sys/i386/compile/D1R74V1 i386
Scanner: Mustek MFS-6000CX
Controller: Symbios 810a
Driver: sym
>Description:
The scanner gets detected, but is inaccessible.
Long time ago, I figured that the sym driver tries to negotiate synchronous
transfer, and the scanner seems to not like this at all, and things
go into some indefinite state. Then I created a crude hack to avoid
this.
With Rel. 7 that crude hack seemed to no longer work. (In fact it
did still work - the real problem was that "camcontrol" is flawed
since Rel. 7, see bin/163053.)
So I started to search for the root cause.
Q: why does the driver try to negotiate synchronous with everybody,
even with devices that have no idea about synch transfers?
And, _if_ it does so deliberately, shouldn't there be a quirk to
switch it off for devices that misbehave?
I finally found the reason: in sym_hipd.c:sym_update_trans() there is
a range-check done for the synchronous parameters.
This range-check seems to disregard the possibility that there could
be *no* synchronous transfer intended, i.e. the parameter could be 0.
>How-To-Repeat:
Activate the DEBUG_NEGO option in sym_hipd.c
> #define DEBUG_FLAGS (0x0200)
and watch the negotiation messages.
Thats an U160 disk on a 1010-33 controller - I guess thats how it is
supposed to work:
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): wide msgin: 1-2-3-1.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): wide msgout: 1-2-3-1.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): sync msgin: 1-3-1-a-7f.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): sync msgout: 1-3-1-a-1f.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): wide msgout: 1-2-3-1.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): wide msgin: 1-2-3-1.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): sync msgout: 1-3-1-a-1f.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): sync msgin: 1-3-1-a-1f.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): ppr msgout: 1-6-4-9-0-3e-1-2.
Dec 3 17:49:51 <kern.crit> disp kernel: (probe2:sym0:0:2:0): ppr msgin: 1-6-4-9-0-3e-1-2.
Thats a CDrom on a 810a controller - it can do sync transfers, but we
see the controller starts negotiation with someway strange parameters:
Dec 4 06:01:41 <kern.crit> disp kernel: (probe35:sym2:0:5:0): sync msgout: 1-3-1-19-0.
Dec 4 06:01:41 <kern.crit> disp kernel: (probe35:sym2:0:5:0): sync msgin: 1-3-1-19-0.
Dec 4 06:01:41 <kern.crit> disp kernel: (probe35:sym2:0:5:0): sync msgout: 1-3-1-19-8.
Dec 4 06:01:41 <kern.crit> disp kernel: (probe35:sym2:0:5:0): sync msgin: 1-3-1-19-8.
Dec 4 06:01:41 <kern.crit> disp kernel: (probe35:sym2:0:5:0): sync msgout: 1-3-1-19-8.
Dec 4 06:01:41 <kern.crit> disp kernel: (probe35:sym2:0:5:0): sync msgin: 1-3-1-19-8.
Dec 4 06:01:41 <kern.crit> disp kernel: (probe35:sym2:0:5:0): sync msgout: 1-3-1-19-8.
Dec 4 06:01:41 <kern.crit> disp kernel: (probe35:sym2:0:5:0): sync msgin: 1-3-1-19-8.
Thats a scanner - it has no idea about sync transfers, and seems to
ignore the messages:
Dec 4 02:28:58 <kern.crit> disp kernel: (probe1:sym2:0:1:0): sync msgout: 1-3-1-19-0.
Dec 4 02:28:59 <kern.crit> disp kernel: (probe1:sym2:0:1:0): sync msgout: 1-3-1-19-0.
Dec 4 02:29:16 <kern.crit> disp kernel: (probe1:sym2:0:1:0): sync msgout: 1-3-1-19-0.
Dec 4 02:29:45 <kern.crit> disp kernel: (probe1:sym2:0:1:0): sync msgout: 1-3-1-19-0.
Dec 4 02:30:03 <kern.crit> disp kernel: (probe1:sym2:0:1:0): sync msgout: 1-3-1-19-0.
Dec 4 02:30:05 <kern.crit> disp kernel: (probe1:sym2:0:1:0): sync msgout: 1-3-1-19-0.
Dec 4 02:30:05 <kern.crit> disp kernel: (probe1:sym2:0:1:0): sync msgout: 1-3-1-19-0.
Dec 4 02:30:18 <kern.crit> disp kernel: (probe1:sym2:0:1:0): sync msgout: 1-3-1-19-0.
Thats the Mustek scanner - it seems to be pissed off and will not
work - while the driver insists and retries:
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): sync msgout: 1-3-1-19-0.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): phase change 6-7 6 at 01ab17a8 resid=4.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): sync msgout: 1-3-1-19-0.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): phase change 6-7 6 at 01ab17a8 resid=4.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): sync msgout: 1-3-1-19-0.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): phase change 6-7 6 at 01ab17a8 resid=4.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): sync msgout: 1-3-1-19-0.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): phase change 6-7 6 at 01ab17a8 resid=4.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): sync msgout: 1-3-1-19-0.
Dec 3 18:24:42 <kern.crit> disp kernel: (probe33:sym2:0:3:0): phase change 6-7 6 at 01ab17a8 resid=4.
...
>Fix:
*** sys/dev/sym/sym_hipd.c.orig Thu Dec 1 05:26:17 2011
--- sys/dev/sym/sym_hipd.c Sun Dec 4 06:35:57 2011
***************
*** 8291,8297 ****
*/
if (tip->width > SYM_SETUP_MAX_WIDE) tip->width = SYM_SETUP_MAX_WIDE;
if (tip->offset > SYM_SETUP_MAX_OFFS) tip->offset = SYM_SETUP_MAX_OFFS;
! if (tip->period < SYM_SETUP_MIN_SYNC) tip->period = SYM_SETUP_MIN_SYNC;
/*
* Scale against actual controller BUS width.
--- 8291,8298 ----
*/
if (tip->width > SYM_SETUP_MAX_WIDE) tip->width = SYM_SETUP_MAX_WIDE;
if (tip->offset > SYM_SETUP_MAX_OFFS) tip->offset = SYM_SETUP_MAX_OFFS;
! if (tip->period < SYM_SETUP_MIN_SYNC && tip->period != 0)
! tip->period = SYM_SETUP_MIN_SYNC;
/*
* Scale against actual controller BUS width.
***************
*** 8319,8325 ****
tip->offset = np->maxoffs_dt;
}
else {
! if (tip->period < np->minsync)
tip->period = np->minsync;
if (tip->period > np->maxsync)
tip->period = np->maxsync;
--- 8320,8326 ----
tip->offset = np->maxoffs_dt;
}
else {
! if (tip->period < np->minsync && tip->period != 0)
tip->period = np->minsync;
if (tip->period > np->maxsync)
tip->period = np->maxsync;
>Release-Note:
>Audit-Trail:
>Unformatted:
More information about the freebsd-bugs
mailing list